Deformable registration of 4D-CT lung image using landmark

Landmark based image registration algorithms have been applied to medical image analysis widely. While estimation of the correspondence between landmarks among images are the most important and difficult problem. In this paper, novel landmarks based image registration method combining local tracking and global adjustment of landmarks in 4D-CT lung images is proposed. In order to accurately establish the corresponding landmarks in images at different phases of 4D-CT, we introduce spatial information to structure tensor and use structure tensor to track landmarks in images at different phases. The prior knowledge of respiratory motion is utilized to limit the search area of corresponding landmarks. Further, a cost function measuring similarity of images, bending energy of transformations and spatio-temporal smoothness of trajectories of landmarks along respiratory phases is given. We update the tracked results of landmarks by minimizing a cost function. After the optimization, the correspondence between landmarks is improved and the trajectories of landmarks are enforced to be more smooth, thus, more accurate registration results can be achieved. Experimental results of public lung dataset demonstrate the effectiveness of the proposed method.